This paper reviews various methods for evaluating image segmentation algorithms, classifying them into three main groups: analytical, empirical goodness, and empirical discrepancy. Each group has unique characteristics and advantages. Analytical methods directly assess algorithms by analyzing their principles and properties, while empirical methods indirectly judge algorithms by comparing segmented images to reference images. Empirical methods are further divided into goodness methods, which measure desirable properties of segmented images, and discrepancy methods, which compare segmented and reference images to quantify differences. The paper discusses the strengths and limitations of each method group, emphasizing the importance of considering application-specific knowledge and the need for objective and quantitative evaluation. An experimental comparison of several empirical methods reveals that discrepancy methods, particularly those based on feature values and mis-segmented pixels, are more effective than goodness methods. The study also highlights the need for domain-dependent knowledge and the importance of combining multiple evaluation methods to comprehensively assess segmentation algorithms. The paper concludes by emphasizing the necessity of further research and efforts in segmentation evaluation to improve the performance of existing algorithms and develop new, powerful methods.This paper reviews various methods for evaluating image segmentation algorithms, classifying them into three main groups: analytical, empirical goodness, and empirical discrepancy. Each group has unique characteristics and advantages. Analytical methods directly assess algorithms by analyzing their principles and properties, while empirical methods indirectly judge algorithms by comparing segmented images to reference images. Empirical methods are further divided into goodness methods, which measure desirable properties of segmented images, and discrepancy methods, which compare segmented and reference images to quantify differences. The paper discusses the strengths and limitations of each method group, emphasizing the importance of considering application-specific knowledge and the need for objective and quantitative evaluation. An experimental comparison of several empirical methods reveals that discrepancy methods, particularly those based on feature values and mis-segmented pixels, are more effective than goodness methods. The study also highlights the need for domain-dependent knowledge and the importance of combining multiple evaluation methods to comprehensively assess segmentation algorithms. The paper concludes by emphasizing the necessity of further research and efforts in segmentation evaluation to improve the performance of existing algorithms and develop new, powerful methods.